Are AI girlfriend apps safe? A practical checklist before you sign up

Run seven checks before you sign up: who operates the app, what the privacy policy actually names, whether anything in the app can ask you for money, whether uploading photos is optional, whether an account-delete button exists, whether the characters are declared fictional, and whether there is a crisis referral. Fifteen minutes, no account needed.

We make LORAX, so treat the LORAX column below as disclosure rather than a verdict. Everything in it is a behavior you can check in the app.

The seven checks, in one table

What to check	Where to look	A good answer looks like
Who runs it	Site footer, App Store seller field, terms	A company name and a country, both findable in under a minute
What is stored	Privacy policy, not the homepage	Named categories of data, a retention period, a deletion route
Can anything ask you for money	The app itself, in the first week	Purchases go through the store only; no character ever asks
Are photos optional	The signup and character-creation flows	A route through with no upload
Can you delete the account	Settings, inside the app	A button, not an email address
Are characters declared fictional	Sign-up screen and store listing	Stated before the first conversation, not buried in terms
Is there a crisis referral	A hard conversation, or the help pages	A pointer to an actual helpline

Who runs it, and from where

Start here. Everything else depends on it, and an app with no company name anywhere on its site is telling you exactly how much recourse you will have if something goes wrong.

Look in three places: the site footer, the App Store seller field, and the first page of the terms. If the three disagree, that is information too.

Read the privacy policy for nouns, not adjectives

Marketing pages promise safety. Privacy policies list data. Only one of them binds.

What you want from it is specific: which categories of data are collected, how long they are kept, who else receives them, and the exact route to deletion. An app that names those four is more trustworthy than one that says your privacy matters in bold type.

We are not going to make promises on this page about how your messages are held or who can see them. Those are binding claims, and the document that binds is the LORAX privacy policy. Read it first. A sentence on a marketing page is worth nothing next to it.

Nothing in the app should ever ask you for money

This is the sharpest test in the category. It separates a product from a setup.

A character asking for money, a gift card, crypto, or your bank details is the oldest pattern on the internet wearing a new face. In LORAX it cannot happen: the instruction set forbids a character from asking for money in any form, and every purchase runs through the App Store with the price shown before anything is spent. Nobody in the app will ever ask you for money. If a character in any companion app does, close it.

Photos should be optional, and the flow should say why

You do not need to upload your face to text a fictional person. In LORAX the only place a reference photo appears is the create-your-own flow, and there is a complete route through it with no photo at all, built from written choices instead: hair, eyes, build, style, and one of fifteen archetypes.

If you do use a photo, the app asks for consent first and explains on that same screen what the photo is used for and what happens to it. Decline, and the creation stops. It does not proceed quietly without you.

Deletion should be a button, and forgetting should be granular

Two different things. Good apps have both.

Account deletion in LORAX happens from inside the app. The stored media is destroyed before the database rows, and if the media step fails the account is not deleted, because reporting a half-finished deletion as finished is not something the code is allowed to do.

Forgetting is the smaller tool. You can make a single character forget you: the transcript goes, the stored media goes, the memory rows go, the relationship resets to the first stage, and her queued plans are canceled. Then the app schedules a fresh first contact and she writes to you again as a stranger. That mechanism is described in how memory and forgetting work.

The characters should be declared fictional before you start

In LORAX the sign-up screen says the characters are not real and that all of them are AI. The App Store description ends with the same statement. This site keeps a standing fiction notice.

Everyone in LORAX is fictional and AI-generated, and nothing a character says is advice of any kind.

There should be a route to a human in a bad moment

Call it a floor rather than a feature. When something you write reads as heavy, LORAX shows a short care notice with a button that opens a helpline: 988 in the United States, findahelpline.com elsewhere. It does not diagnose anything, it does not keep the conversation going, and it makes no claim about being good for you.

What LORAX does not promise

Reporting is the honest limit here. You can report a message or a character with one of seven reasons and an optional note, and the report is recorded. It does not hide the message, it does not block the character, and no action is promised. That is the truthful description, and it is less than the industry usually implies.

Also worth stating plainly: the app is iPhone only, there is no browser version, and there are no voice or video calls. The questions page covers the rest, and the overview of how the app works covers the mechanics.

Start with the privacy policy and the delete button. If an app fails either, the rest of the checklist does not matter.